Building Jist: An AI-Powered Notification Summarizer with GitHub Copilot CLI
Source: Dev.to
Jist is an open‑source Android application that intercepts notifications from apps such as WhatsApp, Telegram, Gmail, Slack, and more. It batches notifications per contact, group, or thread and uses LLM APIs to generate concise, intelligent summaries.
Core Features
- 🔔 Notification Capture – Listens to all app notifications via Android’s
NotificationListenerService. - 🤖 Multi‑Model LLM Support – OpenAI, Gemini, Claude, OpenRouter, and any OpenAI‑compatible API.
- ⚙️ Model Selection – Configure your preferred LLM provider and model.
- 📋 Auto Mode – Background summarization of notifications.
- 🔑 Bring Your Own Key – Users provide their own API keys; no backend required.
- 🔒 Privacy First – All processing is on‑device; only notification text is sent to the chosen LLM.
- 🌐 Modern UI – Built with Material Design 3.
Technology Stack
| Component | Technology |
|---|---|
| Language | Kotlin |
| UI | Jetpack Compose, Material Design 3 |
| Database | Room |
Development with GitHub Copilot CLI
Model Strategy
| Model | Best For | Why |
|---|---|---|
| Gemini 3 Pro | UI / Compose code | Excels at visualizing layouts, Material 3 components, and Jetpack Compose patterns. |
| Opus 4.6 | Planning & architecture | Superior reasoning for system design, database schemas, and complex logic. |
| GPT‑5.2 Codex / Sonnet | General development | Balanced for boilerplate, testing, and debugging. |
Skills Utilized
- Android Jetpack Compose – Declarative UI building with modern Compose patterns.
- Material Design 3 – Implementing Material You theming, dynamic colors, and expressive components.
These skills gave context‑aware suggestions specific to Android development, reducing the learning curve significantly.
Model Context Protocol (MCP) Servers
- File System MCP – Navigates the Android project structure and understands the codebase.
- GitHub MCP – Checks issues and manages the development workflow.
- Custom Context Providers – Supplies Android‑specific patterns and best practices.
MCPs enabled Copilot to grasp project context deeply, resulting in more accurate and relevant suggestions.
Benefits Observed
- Rapid Prototyping – Boilerplate code generation let me focus on business logic.
- Learning Jetpack Compose – Quick answers about Material 3 components accelerated my onboarding.
- Architecture Decisions – Opus guided database schema and service architecture design.
- Debugging – Claude helped trace complex code flows when issues arose.
- UI Polish – Gemini’s visual code strengths produced a polished, modern interface.
Quantitative Impact
- Time Saved – Significant reduction in boilerplate coding.
- Quality Improved – Better adherence to Jetpack Compose and Material 3 best practices.
- Learning Accelerated – Fast‑tracked understanding of Android architecture components.
- Creativity Boosted – More focus on unique features rather than syntax hurdles.
Notable Challenges Solved
- Onboarding screen crash – Copilot identified and fixed nested scrollable container constraints.
- Battery optimization across OEMs – Researched and implemented guidance for Samsung, OnePlus, Xiaomi, etc.
- Unified LLM provider interface – Designed a clean factory pattern handling OpenAI, Gemini, Claude, and OpenRouter APIs.
Conclusion
GitHub Copilot CLI transformed the development experience for Jist. Switching between specialized AI models for distinct tasks—rather than a one‑size‑fits‑all approach—unlocked new levels of productivity and code quality. The combination of Copilot’s agentic capabilities, MCP servers, and model selection made this complex Android application achievable in a fraction of the time it would have otherwise required.
GitHub Repository: